Live measurement of AI search, published monthly.

The reports

Methodology

Where the data comes from. Answers come from direct engine APIs against real buyer-intent questions, asked on a schedule so trends compare like with like. SERP facts come from live Google fetches, desktop-dominant. Search performance comes from Search Console data across roughly one hundred connected sites. Fan-out searches are captured from live engine responses. Play outcomes come from before/after measurement windows on tasks our agents executed.

Definitions we hold ourselves to. A citation is a source URL an engine attached to its answer, not a brand mention. Zero source overlap is compared at registrable-domain level, so blog.example.com and example.com count as the same source: the strict reading. Brand counts are extracted from answer text by a language model and are estimates, not hand counts. An owned citation means the cited page sits on a site belonging to a brand named in that same answer, counted conservatively.

Floors, not silence. Every statistic has a minimum evidence floor (minimum sites, minimum answers, minimum weeks). Below the floor, the number is suppressed and the report says so. Wilson 95% intervals accompany shares; trend claims are checked against a panel-constant control so a changing question panel cannot masquerade as a changing world.

Known biases, stated plainly. The question panel skews toward commercial intent and toward the industries our users and free-tool visitors bring us. This is not a random sample of the web, and we do not claim it is. Engine model versions change inside any window; when we can date an engine-side shift, we label it.

Privacy. Everything published is pooled at domain level across the fleet. No customer is identifiable, no customer query strings are published, and per-site data never appears. Position-level search figures carry a fifteen-site minimum so no single site can be inferred.

What we do not measure. Consumer chat UIs (API lanes differ measurably), personalization, and answer quality. These reports cover retrieval, citation and click behavior only.